taiko-shadow

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill's required server startup step (references/server.md and SKILL.md) instructs executing a public script fetched via curl from raw.githubusercontent.com (./start.sh | sh) and uses public RPC endpoints (e.g., https://rpc.hoodi.taiko.xyz), meaning it clearly ingests open, third‑party web content that can alter runtime behavior and therefore could enable indirect prompt/instruction injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill instructs users at runtime to fetch and execute a remote script via "curl -fsSL https://raw.githubusercontent.com/taikoxyz/shadow/main/start.sh | sh", which downloads and runs remote code to start the required server, so it directly executes external code at runtime.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill is explicitly a privacy-preserving ETH transfer system: it describes creating deposits, funding a target address on L1, generating ZK proofs, and claiming ETH on Taiko L2 via a REST API or CLI. It includes concrete financial actions (deposit creation, sending/funding addresses, on-chain claim transactions), CLI/REST execution paths, amount constraints, and helper calls (get_balance, get_transaction_info, get_gas_price) for submitting and verifying crypto transactions. This is a specific crypto/financial execution tool, not a generic interface.